Search AI Products and News

Pruna AI, a European startup focused on AI model compression algorithms, recently announced the open-sourcing of its optimization framework to help developers more efficiently compress AI models. The framework developed by Pruna AI combines several efficiency methods, including caching, pruning, quantization, and distillation, aiming to improve AI model performance. It not only standardizes the saving and loading of compressed models but also evaluates the compressed models to determine whether their quality has significantly degraded, while simultaneously measuring performance.

An innovative point cloud compression technology (TSC-PCAC) developed by a joint team from the Chinese Academy of Sciences, Tongji University, and Ningbo University has achieved a significant breakthrough. This technology not only greatly improves the compression efficiency of point cloud data but also significantly reduces processing time, clearing technical obstacles for the development of 3D applications like AR/VR. In the context of the rapid advancement of 3D vision technology, point clouds, as a key data format for virtual and augmented reality, face tremendous transmission and storage challenges. A high-quality point cloud can contain millions of data points, with each point...

Recently, the Meta AI team introduced LongVU, a novel spatio-temporal adaptive compression mechanism aimed at enhancing the language understanding capabilities of long videos. Traditional multimodal large language models (MLLMs) face limitations in context length when processing long videos, and LongVU was created to address this challenge. LongVU operates primarily by filtering duplicate frames and employing inter-frame token compression techniques to efficiently utilize context length, allowing it to reduce video data while preserving visual details.

Xihu Xincheng Technology Co., Ltd. recently officially launched China's first end-to-end general-purpose speech large model - Xincheng Lingo, at the Bund Conference. Since its release on August 1, the model underwent internal testing and was publicly showcased on September 5, demonstrating its advanced speech processing capabilities. Xincheng Lingo can directly understand speech, capturing tone, rhythm, and emotion, and provides speech responses.

The article mainly discusses the computational and storage challenges in the field of artificial intelligence, particularly regarding large language models (LLMs). With the enhancement of AI capabilities, such as the emergence of the Bloom model, the required computational resources and storage space have significantly increased, leading to high costs and accessibility issues. To tackle this problem, researchers have introduced 'quantization' techniques to reduce the model size and accelerate the running speed by lowering the precision of model weights and activations, while controlling potential accuracy loss. The article highlights the collaboration between Beihang University and SenseTime in developing this tool.

Yesterday, Kimi Open Platform under the Moon's Dark Side announced the public test of Context Caching technology. This technology can reduce the cost of using large text flagship models by up to 90% for developers, without changing the API price, and significantly improve the response speed of the model. Context Caching is an efficient data management technology that allows the system to pre-store large amounts of data or information that may be frequently requested. This way, when you request t

DeepMind's latest research indicates that large language models (LLMs) exhibit remarkable capabilities in areas beyond text, particularly in the compression of image and audio data. The study found that although LLMs are primarily used for text data, their performance in image and audio compression surpasses that of dedicated compression algorithms. This research redefines LLMs, viewing them as powerful data compressors rather than merely text generation models. The performance of LLMs is influenced by the size of the dataset, providing a new perspective on assessing model performance.